How to start a dark kitchen from scratch: the mistakes that destroy margin and the method that protects it

Chapter 1 — What has to be settled before you sign the kitchen lease?

Before you sign anything, you need a menu price that absorbs a 28% platform take rate and still leaves positive contribution margin. That is the calculation, and it happens on a spreadsheet, not on site.

A $10 dish carrying 32% food cost leaves $6.80 in a dining room; that same dish under aggregator retention drops to $3.60, and no growth in order count repairs a structure that started crooked. Demand exists. Profit does not follow automatically. Diego F. Parra insists at Masterestaurant that the take rate belongs in the P&L as a CHANNEL COST from the very first line, never as a temporary commercial discount.

Chapter 2 — Platform concentration decides who sets the price

You are not negotiating with a competitive market of aggregators: you are negotiating with a regional near-monopoly, and that sets the ceiling on what you can ask for. Earnest Analytics measured DoorDash at 60.7% of US delivery at the close of 2024, against 26.1% for Uber Eats and barely 6.3% for Grubhub. In Brazil the asymmetry runs harsher still: a single platform holds the vast majority of online food bookings and processes a monthly order volume no independent operator can match alone. Grab commands 53.9% of Southeast Asia according to Momentum Works (2024). Spain looks more open, with Glovo near 31% and Just Eat near 26% (Ken Research, 2025), and there you genuinely have leverage to argue commission points. Outside that case, asking for a discount from whoever controls six of every ten orders is a wasted conversation.

Chapter 3 — CapEx rises, but variable cost is what kills operations

Almost everyone arrives worried about the upfront investment and ends up buried by the monthly commission. It runs backwards from the story. A 40-square-meter hidden kitchen removes dining room rent, furniture, tableware and half the floor payroll, so start-up investment genuinely falls; in exchange you import a 25% to 30% retention on every unit of revenue, which never declines with volume and never amortizes. An investment gets paid once. A commission gets paid every day, all of them, for as long as the business exists. That contrast explains why operators with high occupancy and a full kitchen still close: they bill plenty and contribute little. Break-even has to be calculated on platform NET revenue, after retention, or the number you are staring at is fiction.

Chapter 4 — Revenue bands: the same model is not the same business

Size changes the recommendation completely, and lumping everything under "dark kitchen" is the most expensive analytical error in the sector. Below $500,000 a year, with two people and a single aggregator, a 28% take rate consumes nearly all operating margin and the only realistic exit is an owned channel with pickup at the door. Between $500,000 and $1 million, corporate plans become negotiable and two virtual brands can run over one hot line. Between $1 and $5 million an in-house order manager and a part-time cost analyst start paying for themselves. The share independents hold of cloud kitchen revenue confirms something uncomfortable: the base of this format is small and mid-sized firms, precisely the tier with the least bargaining power against the platform.

Chapter 5 — Above $5 million the problem stops being margin and becomes brand

Past $5 million a year, a group with a media chef or a large-format themed concept plays a different match: its effective retention falls into the 15% to 20% range because volume opens a master agreement, and the cost threatening it is another one entirely. That operator carries R&D recipe development, a content team and a fixed kitchen payroll that does not collapse in low season. Past the $10 million mark, co-packing for third parties and centralized production appear, with internal transfers between units, which shifts the problem to inventory traceability. Notice the paradox: whoever gets the cheapest channel is whoever needs it least, and whoever pays 30% is the $400,000 kitchen that cannot absorb it. The small tier should not copy that playbook. It should avoid it.

Chapter 6 — What happens if three competitors open inside your delivery polygon

Assume that within six months three more kitchens serve your radius in your category. Orders do not split into four equal parts: the algorithm rewards the best rating and delivery time, so fourth place loses visibility and compensates by buying paid placement, which raises acquisition cost exactly as average ticket falls. That is where the spiral begins. The defense is not cutting price. It is building direct demand —WhatsApp, pickup at the door, corporate meal subscriptions— before the polygon saturates, because afterwards working capital will not stretch to fight for placement and fund operations at the same time.

Chapter 7 — Formalization, jobs and traceability: the reading development banks care about

For a multilateral program officer, every registered hidden kitchen represents declared employment and fiscal traceability in a historically informal sector, and that is the argument that unlocks concessional credit lines. It should not be romanticized. Europe accounts for a much smaller share of the global dark kitchen market than other regions while holding a meaningful share of the delivery app market, an asymmetry pointing to a regulated market where the kitchen without a dining room grows less and survives better. Latin America runs the opposite way: it grows fast and survives worse. A serious support program should not finance equipment; it should finance working capital conditioned on the applicant showing break-even net of retention. That is a one-page document, and it screens better than any collateral requirement.

Chapter 8 — The first operating decision: short menu, one hot line

Start with eight dishes, not twenty-four. A short menu sustains theoretical food cost below 32%, which in this format is a hard ceiling rather than a target, because it lets you buy volume across few inputs and eliminates slow-rotation waste. Every added item multiplies inventory, spoilage and dispatch time, and dispatch time is precisely the metric the aggregator uses to rank the list where the customer chooses. Automation comes later: Grand View Research assigns North America 40.8% of the kitchen robotics market, a signal that this spending still lives where labor cost justifies it, not in a 40-meter kitchen in Bogotá or Lima. Your opening advantage is arithmetic, not technological. Run break-even on net revenue this week, and if the number fails with eight dishes, it fails with twenty-four.

Chapter 9 — Differences that decide whether the operation survives year two

The first difference is arithmetic and not open to debate: a dish sold at USD 10 in the dining room at 32% food cost leaves USD 6.80 of gross margin; that same dish on a platform keeping 28% leaves USD 3.60, barely half, and no amount of volume repairs a structure that starts bent. Where a seated restaurant competes for physical location, a dark kitchen competes for placement in an algorithmic list. With a handful of platforms controlling most of the delivery market in both the US and Brazil, what you negotiate is not shelf space: it is visibility inside a recommendation engine you do not control. CapEx behaves opposite to most intuitions. A kitchen without a dining room saves on furniture, façade and service staff, yet demands spending on thermal packaging, dispatch systems and production capacity concentrated in two two-hour peaks — the saving is real, and smaller than the sector's pitch suggests.

Chapter 10 — Differences that decide whether the operation survives year two — in practice

Labor risk changes shape. No servers means no tips, and without tips the kitchen worker's full income has to come out of wages, which raises formal payroll cost and pushes toward informality — a point the decent work agenda under SDG 8 forces you to confront when financing these projects. A virtual brand has a shorter life cycle than a restaurant. One kitchen can host three or four brands and retire whichever underperforms within ninety days, something impossible with a physical site, and that flexibility is the format's genuine competitive edge — provided the operator has the discipline to kill what fails to deliver margin. Waste behaves differently: with no buffet, no display shrink and production strictly to order, food loss drops — but so does traceability, because nobody returns a cold dish, they simply leave a one-star review and never come back.